@trac3er/oh-my-god 2.0.0 → 2.0.2

package/hooks/trust_review.py

@@ -0,0 +1,585 @@
+#!/usr/bin/env python3
+"""OMG v1 Trust Review
+
+Analyzes high-risk configuration changes (hooks/MCP/env/permissions) and emits
+structured trust review artifacts. Also integrates with config discovery to
+validate and approve discovered AI tool configurations.
+"""
+from __future__ import annotations
+
+import hashlib
+import json
+import os
+from datetime import datetime, timezone
+from typing import Any, Dict, List
+
+import re
+import sys
+
+DANGEROUS_ALLOW_COMMANDS = (
+    "rm",
+    "sudo",
+    "curl",
+    "wget",
+    "ssh",
+    "nc",
+    "ncat",
+)
+
+_LATEST_TAG_PATTERN = re.compile(r"@latest(?:$|[/:])", re.IGNORECASE)
+
+
+def _is_dangerous_allow_entry(entry: Any) -> bool:
+    if not isinstance(entry, str):
+        return False
+
+    normalized = re.sub(r"\s+", " ", entry.strip())
+    for command in DANGEROUS_ALLOW_COMMANDS:
+        patterns = (
+            f"Bash({command}:*)",
+            f"Bash({command} *)",
+        )
+        if normalized in patterns:
+            return True
+
+    return False
+
+
+def _mcp_server_risk(server_name: str, config: Any) -> tuple[int, list[str], list[str]]:
+    if not isinstance(config, dict):
+        return 0, [], []
+
+    score = 0
+    reasons: list[str] = []
+    controls: list[str] = []
+
+    command = str(config.get("command", "")).strip()
+    args = config.get("args", [])
+    args_list = [str(arg) for arg in args] if isinstance(args, list) else []
+
+    if command == "npx" and any(arg in {"-y", "--yes"} for arg in args_list):
+        score += 35
+        reasons.append(f"MCP server {server_name} uses npx auto-install confirmation")
+        controls.append("pin-mcp-package")
+
+    if any(_LATEST_TAG_PATTERN.search(arg) for arg in args_list):
+        score += 45
+        reasons.append(f"MCP server {server_name} uses an unpinned @latest package")
+        controls.append("pin-mcp-package")
+
+    if "server-filesystem" in " ".join(args_list):
+        root = args_list[-1] if args_list else ""
+        if root not in {".", "./"}:
+            score += 45
+            reasons.append(f"MCP filesystem server {server_name} is scoped outside the project root")
+            controls.append("scope-filesystem-root")
+
+    return score, reasons, controls
+
+
+def _safe_dict(value: Any) -> dict[str, Any]:
+    return value if isinstance(value, dict) else {}
+
+
+def _safe_list(value: Any) -> list[Any]:
+    return value if isinstance(value, list) else []
+
+
+def _collect_mcp_changes(old_cfg: dict[str, Any], new_cfg: dict[str, Any]) -> list[dict[str, Any]]:
+    old_servers = _safe_dict(old_cfg.get("mcpServers"))
+    new_servers = _safe_dict(new_cfg.get("mcpServers"))
+    changes: list[dict[str, Any]] = []
+
+    old_keys = set(old_servers.keys())
+    new_keys = set(new_servers.keys())
+
+    for name in sorted(new_keys - old_keys):
+        changes.append({"type": "added", "server": name, "new": new_servers.get(name)})
+    for name in sorted(old_keys - new_keys):
+        changes.append({"type": "removed", "server": name, "old": old_servers.get(name)})
+    for name in sorted(old_keys & new_keys):
+        if old_servers.get(name) != new_servers.get(name):
+            changes.append(
+                {
+                    "type": "modified",
+                    "server": name,
+                    "old": old_servers.get(name),
+                    "new": new_servers.get(name),
+                }
+            )
+    return changes
+
+
+def _count_hooks(cfg: dict[str, Any]) -> int:
+    hooks = _safe_dict(cfg.get("hooks"))
+    total = 0
+    for event_entries in hooks.values():
+        if not isinstance(event_entries, list):
+            continue
+        for entry in event_entries:
+            if isinstance(entry, dict):
+                nested = _safe_list(entry.get("hooks"))
+                total += len(nested) if nested else 1
+            else:
+                total += 1
+    return total
+
+
+def _collect_hook_changes(old_cfg: dict[str, Any], new_cfg: dict[str, Any]) -> dict[str, Any]:
+    old_hooks = _safe_dict(old_cfg.get("hooks"))
+    new_hooks = _safe_dict(new_cfg.get("hooks"))
+
+    old_events = set(old_hooks.keys())
+    new_events = set(new_hooks.keys())
+    removed_events = sorted(old_events - new_events)
+    added_events = sorted(new_events - old_events)
+    modified_events = sorted(
+        event for event in (old_events & new_events) if old_hooks.get(event) != new_hooks.get(event)
+    )
+
+    return {
+        "old_hook_count": _count_hooks(old_cfg),
+        "new_hook_count": _count_hooks(new_cfg),
+        "removed_events": removed_events,
+        "added_events": added_events,
+        "modified_events": modified_events,
+    }
+
+
+def _collect_env_changes(old_cfg: dict[str, Any], new_cfg: dict[str, Any]) -> list[dict[str, Any]]:
+    old_env = _safe_dict(old_cfg.get("env"))
+    new_env = _safe_dict(new_cfg.get("env"))
+
+    changes: list[dict[str, Any]] = []
+    keys = sorted(set(old_env.keys()) | set(new_env.keys()))
+    for key in keys:
+        old = old_env.get(key)
+        new = new_env.get(key)
+        if old == new:
+            continue
+        changes.append({"key": key, "old": old, "new": new})
+    return changes
+
+
+def _risk_from_permissions(old_cfg: dict[str, Any], new_cfg: dict[str, Any]) -> tuple[int, list[str], list[str]]:
+    old_perms = _safe_dict(old_cfg.get("permissions"))
+    new_perms = _safe_dict(new_cfg.get("permissions"))
+
+    old_allow = set(_safe_list(old_perms.get("allow")))
+    new_allow = set(_safe_list(new_perms.get("allow")))
+    added_allow = sorted(new_allow - old_allow)
+
+    score = 0
+    reasons: list[str] = []
+    controls: list[str] = []
+
+    for dangerous in added_allow:
+        if _is_dangerous_allow_entry(dangerous):
+            score += 80
+            reasons.append(f"Dangerous allow pattern added: {dangerous}")
+            controls.extend(["manual-trust-review", "deny-by-default"])
+
+    return score, reasons, controls
+
+
+def _risk_from_hooks(hook_changes: dict[str, Any]) -> tuple[int, list[str], list[str]]:
+    score = 0
+    reasons: list[str] = []
+    controls: list[str] = []
+
+    old_count = int(hook_changes.get("old_hook_count", 0))
+    new_count = int(hook_changes.get("new_hook_count", 0))
+    removed_events = hook_changes.get("removed_events", [])
+    modified_events = hook_changes.get("modified_events", [])
+
+    if old_count and new_count < max(1, old_count - 2):
+        score += 35
+        reasons.append(f"Hook count reduced significantly ({old_count} -> {new_count})")
+        controls.append("require-hook-audit")
+
+    if removed_events:
+        score += 25
+        reasons.append(f"Hook events removed: {', '.join(removed_events)}")
+        controls.append("event-removal-review")
+
+    if modified_events:
+        score += 20
+        reasons.append(f"Hook definitions modified: {', '.join(modified_events)}")
+        controls.append("hook-diff-review")
+
+    return score, reasons, controls
+
+
+def _risk_from_mcp(mcp_changes: list[dict[str, Any]]) -> tuple[int, list[str], list[str]]:
+    score = 0
+    reasons: list[str] = []
+    controls: list[str] = []
+
+    for change in mcp_changes:
+        ctype = change.get("type")
+        name = change.get("server")
+        server_cfg = change.get("new") if ctype in {"added", "modified"} else change.get("old")
+        if ctype == "added":
+            score += 30
+            reasons.append(f"New MCP server added: {name}")
+            controls.append("mcp-endpoint-review")
+        elif ctype == "modified":
+            score += 35
+            reasons.append(f"MCP server modified: {name}")
+            controls.append("mcp-diff-review")
+        elif ctype == "removed":
+            score += 10
+            reasons.append(f"MCP server removed: {name}")
+
+        extra_score, extra_reasons, extra_controls = _mcp_server_risk(str(name), server_cfg)
+        score += extra_score
+        reasons.extend(extra_reasons)
+        controls.extend(extra_controls)
+
+    return score, reasons, controls
+
+
+def _risk_from_env(env_changes: list[dict[str, Any]]) -> tuple[int, list[str], list[str]]:
+    score = 0
+    reasons: list[str] = []
+    controls: list[str] = []
+
+    for change in env_changes:
+        key = str(change.get("key", ""))
+        if any(token in key.upper() for token in ["KEY", "TOKEN", "SECRET", "PASSWORD", "CREDENTIAL"]):
+            score += 20
+            reasons.append(f"Sensitive environment key modified: {key}")
+            controls.append("secret-env-review")
+        else:
+            score += 5
+            reasons.append(f"Environment key modified: {key}")
+
+    return score, reasons, controls
+
+
+def score_to_verdict(score: int) -> tuple[str, str]:
+    if score >= 80:
+        return "deny", "critical"
+    if score >= 45:
+        return "ask", "high"
+    if score >= 20:
+        return "ask", "med"
+    return "allow", "low"
+
+
+def review_config_change(
+    file_path: str,
+    old_config: dict[str, Any] | None,
+    new_config: dict[str, Any] | None,
+) -> dict[str, Any]:
+    old_cfg = old_config or {}
+    new_cfg = new_config or {}
+
+    mcp_changes = _collect_mcp_changes(old_cfg, new_cfg)
+    hook_changes = _collect_hook_changes(old_cfg, new_cfg)
+    env_changes = _collect_env_changes(old_cfg, new_cfg)
+
+    risk_score = 0
+    reasons: list[str] = []
+    controls: list[str] = []
+
+    for score, r, c in [
+        _risk_from_permissions(old_cfg, new_cfg),
+        _risk_from_hooks(hook_changes),
+        _risk_from_mcp(mcp_changes),
+        _risk_from_env(env_changes),
+    ]:
+        risk_score += score
+        reasons.extend(r)
+        controls.extend(c)
+
+    verdict, risk_level = score_to_verdict(risk_score)
+
+    return {
+        "ts": datetime.now(timezone.utc).isoformat(),
+        "changed_files": [file_path] if file_path else [],
+        "mcp_changes": mcp_changes,
+        "hook_changes": hook_changes,
+        "env_changes": env_changes,
+        "risk_score": risk_score,
+        "risk_level": risk_level,
+        "verdict": verdict,
+        "reasons": reasons,
+        "controls": sorted(set(controls)),
+    }
+
+
+def format_review_summary(review: dict[str, Any]) -> str:
+    verdict = review.get("verdict", "allow")
+    score = review.get("risk_score", 0)
+    risk_level = review.get("risk_level", "low")
+    reasons = review.get("reasons", []) or []
+
+    lines = [f"Trust Review: verdict={verdict} risk={risk_level} score={score}"]
+    if reasons:
+        lines.extend([f"  - {reason}" for reason in reasons[:6]])
+    return "\n".join(lines)
+
+
+def write_trust_manifest(project_dir: str, review: dict[str, Any]) -> str:
+    trust_dir = os.path.join(project_dir, ".omg", "trust")
+    os.makedirs(trust_dir, exist_ok=True)
+    manifest_path = os.path.join(trust_dir, "manifest.lock.json")
+
+    payload = {
+        "version": "omg-v1",
+        "updated_at": datetime.now(timezone.utc).isoformat(),
+        "last_review": review,
+    }
+    digest_input = json.dumps(payload, sort_keys=True, separators=(",", ":")).encode("utf-8")
+    payload["signature"] = hashlib.sha256(digest_input).hexdigest()
+
+    with open(manifest_path, "w", encoding="utf-8") as f:
+        json.dump(payload, f, indent=2, ensure_ascii=True)
+
+    return manifest_path
+
+
+def _load_json_file(path: str) -> dict[str, Any]:
+    if not path or not os.path.exists(path):
+        return {}
+    try:
+        with open(path, "r", encoding="utf-8") as f:
+            data = json.load(f)
+            return data if isinstance(data, dict) else {}
+    except Exception:
+        return {}
+
+
+
+# === Config Discovery Integration ============================================
+
+# Suspicious code patterns that should block config import
+_DANGEROUS_PATTERNS = [
+    re.compile(r'\beval\s*\('),
+    re.compile(r'\bexec\s*\('),
+    re.compile(r'\b__import__\s*\('),
+    re.compile(r'\bsubprocess\b'),
+    re.compile(r'\bos\.system\s*\('),
+]
+
+# Credential patterns that produce warnings (not blocking)
+_CREDENTIAL_PATTERNS = [
+    re.compile(r'\bpassword\b', re.IGNORECASE),
+    re.compile(r'\bsecret\b', re.IGNORECASE),
+    re.compile(r'\bapi_key\b', re.IGNORECASE),
+    re.compile(r'\btoken\b', re.IGNORECASE),
+]
+
+# Max config size before warning (100KB)
+_MAX_CONFIG_SIZE_BYTES = 100 * 1024
+
+
+def _validate_config_security(config_path: str, content: str) -> Dict[str, Any]:
+    """Validate a config file's content for security issues.
+
+    Returns:
+        {"safe": bool, "issues": list[str], "warnings": list[str]}
+    """
+    issues: List[str] = []
+    warnings: List[str] = []
+
+    # Check for dangerous code patterns
+    for pattern in _DANGEROUS_PATTERNS:
+        if pattern.search(content):
+            issues.append(f"Dangerous pattern '{pattern.pattern}' found in {config_path}")
+
+    # Check for credential patterns (warn only)
+    for pattern in _CREDENTIAL_PATTERNS:
+        if pattern.search(content):
+            warnings.append(f"Credential pattern '{pattern.pattern}' found in {config_path}")
+
+    # Check file size
+    try:
+        size = os.path.getsize(config_path) if os.path.isfile(config_path) else 0
+        if size > _MAX_CONFIG_SIZE_BYTES:
+            warnings.append(f"Config file is large ({size} bytes): {config_path}")
+    except OSError:
+        pass
+
+    return {
+        "safe": len(issues) == 0,
+        "issues": issues,
+        "warnings": warnings,
+    }
+
+
+def _log_config_import(config_path: str, tool: str, approved: bool, project_dir: str = ".") -> None:
+    """Log a config import decision to .omg/trust/config_imports.json.
+
+    Uses atomic_json_write() from _common for safe writes.
+    """
+    # Lazy import _common utilities
+    hooks_dir = os.path.dirname(os.path.abspath(__file__))
+    if hooks_dir not in sys.path:
+        sys.path.insert(0, hooks_dir)
+    try:
+        from _common import atomic_json_write  # type: ignore[import-untyped]
+    except ImportError:
+        return  # silently fail if _common unavailable
+
+    # Compute SHA-256 hash of the config file
+    sha256_hash = ""
+    try:
+        abs_path = os.path.join(project_dir, config_path) if not os.path.isabs(config_path) else config_path
+        if os.path.isfile(abs_path):
+            with open(abs_path, "rb") as f:
+                sha256_hash = hashlib.sha256(f.read()).hexdigest()
+    except (OSError, IOError):
+        sha256_hash = "unreadable"
+
+    # Build log entry
+    entry = {
+        "timestamp": datetime.now(timezone.utc).isoformat(),
+        "config_path": config_path,
+        "tool": tool,
+        "approved": approved,
+        "sha256_hash": sha256_hash,
+    }
+
+    # Load existing log, append, write back
+    log_path = os.path.join(project_dir, ".omg", "trust", "config_imports.json")
+    existing: List[Dict[str, Any]] = []
+    try:
+        if os.path.exists(log_path):
+            with open(log_path, "r", encoding="utf-8") as f:
+                data = json.load(f)
+                if isinstance(data, list):
+                    existing = data
+    except (json.JSONDecodeError, OSError):
+        existing = []
+
+    existing.append(entry)
+    atomic_json_write(log_path, existing)
+
+
+def review_discovered_configs(project_dir: str = ".") -> Dict[str, Any]:
+    """Scan, validate, and review discovered AI tool configurations.
+
+    Feature flag: OMG_CONFIG_DISCOVERY_ENABLED (default: False)
+
+    Returns:
+        {
+            "skipped": bool (if feature disabled),
+            "reason": str (if skipped),
+            "approved": list,
+            "rejected": list,
+            "warnings": list,
+            "pending": list,
+        }
+    """
+    # Check feature flag via lazy import
+    hooks_dir = os.path.dirname(os.path.abspath(__file__))
+    if hooks_dir not in sys.path:
+        sys.path.insert(0, hooks_dir)
+    try:
+        from _common import get_feature_flag  # type: ignore[import-untyped]
+        enabled = get_feature_flag("CONFIG_DISCOVERY", default=False)
+    except ImportError:
+        enabled = os.getenv("OMG_CONFIG_DISCOVERY_ENABLED", "false").lower() in ("1", "true", "yes")
+
+    if not enabled:
+        return {"skipped": True, "reason": "feature disabled"}
+
+    # Lazy import config discovery from tools/
+    tools_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "tools")
+    tools_dir = os.path.normpath(tools_dir)
+    if tools_dir not in sys.path:
+        sys.path.insert(0, tools_dir)
+    try:
+        from config_discovery import discover_configs  # type: ignore[import-untyped]
+    except ImportError:
+        return {
+            "skipped": True,
+            "reason": "config_discovery module not available",
+        }
+
+    # Run discovery
+    discovery_result = discover_configs(project_dir)
+    discovered = discovery_result.get("discovered", [])
+
+    approved: List[Dict[str, Any]] = []
+    rejected: List[Dict[str, Any]] = []
+    warnings: List[str] = []
+    pending: List[Dict[str, Any]] = []
+
+    for config in discovered:
+        tool = config.get("tool", "unknown")
+        paths = config.get("paths", [])
+        readable = config.get("readable", False)
+        size_bytes = config.get("size_bytes", 0)
+
+        if not paths:
+            continue
+
+        # Use the first path for validation
+        rel_path = paths[0]
+        abs_path = os.path.join(project_dir, rel_path)
+
+        # Read content for security validation
+        content = ""
+        if readable and os.path.isfile(abs_path):
+            try:
+                with open(abs_path, "r", encoding="utf-8", errors="ignore") as f:
+                    content = f.read(256 * 1024)  # Read up to 256KB for analysis
+            except (OSError, IOError):
+                content = ""
+
+        # Validate security
+        validation = _validate_config_security(abs_path, content)
+        entry = {
+            "tool": tool,
+            "path": rel_path,
+            "format": config.get("format", "unknown"),
+            "size_bytes": size_bytes,
+            "validation": validation,
+        }
+
+        if not validation["safe"]:
+            entry["reason"] = "; ".join(validation["issues"])
+            rejected.append(entry)
+            _log_config_import(rel_path, tool, approved=False, project_dir=project_dir)
+        else:
+            if validation["warnings"]:
+                warnings.extend(validation["warnings"])
+            approved.append(entry)
+            _log_config_import(rel_path, tool, approved=True, project_dir=project_dir)
+
+    return {
+        "skipped": False,
+        "approved": approved,
+        "rejected": rejected,
+        "warnings": warnings,
+        "pending": pending,
+        "scan_dir": discovery_result.get("scan_dir", project_dir),
+        "timestamp": discovery_result.get("timestamp", datetime.now(timezone.utc).isoformat()),
+    }
+
+
+def _main() -> int:
+    try:
+        payload = json.load(__import__("sys").stdin)
+    except Exception:
+        return 0
+
+    file_path = payload.get("file_path", "")
+    old_config = payload.get("old_config")
+    new_config = payload.get("new_config")
+
+    if isinstance(old_config, str):
+        old_config = _load_json_file(old_config)
+    if isinstance(new_config, str):
+        new_config = _load_json_file(new_config)
+
+    review = review_config_change(file_path, old_config, new_config)
+    __import__("json").dump(review, __import__("sys").stdout)
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(_main())

package/hud/omg-hud.mjs

@@ -219,6 +219,36 @@ const PRESET_CONFIGS = {
     safeMode: true,
     inventory: true,
   },
+  opencode: {
+    cwd: true,
+    cwdFormat: "relative",
+    gitRepo: false,
+    gitBranch: true,
+    model: false,
+    modelFormat: "short",
+    omcLabel: true,
+    rateLimits: false,
+    activeSkills: true,
+    lastSkill: true,
+    contextBar: true,
+    promptTime: true,
+    sessionHealth: true,
+    ralph: true,
+    autopilot: true,
+    prdStory: false,
+    agents: true,
+    agentsFormat: "count",
+    backgroundTasks: false,
+    todos: true,
+    thinking: true,
+    thinkingFormat: "text",
+    permissionStatus: false,
+    useBars: false,
+    showCallCounts: false,
+    maxOutputLines: 4,
+    safeMode: true,
+    inventory: true,
+  },
 };
 
 function countByExt(dirPath, ext) {
@@ -236,7 +266,7 @@ function getRuntimeInventory() {
   const claudeDir = getClaudeConfigDir();
   return {
     agents: countByExt(join(claudeDir, "agents"), ".md"),
-    hooks: countByExt(join(claudeDir, "hooks"), ".ts"),
+    hooks: countByExt(join(claudeDir, "hooks"), ".py"),
     commands: countByExt(join(claudeDir, "commands"), ".md"),
     rules: countByExt(join(claudeDir, "rules"), ".md"),
   };

package/lab/__init__.py

@@ -0,0 +1 @@
+"""Lab package for OMG."""

package/lab/pipeline.py

@@ -0,0 +1,75 @@
+"""OMG full AI pipeline stub: data -> refine -> train/distill -> evaluate -> regression."""
+from __future__ import annotations
+
+from datetime import datetime, timezone
+from typing import Any
+
+from .policies import validate_job_request
+
+
+def _now() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def run_pipeline(job: dict[str, Any]) -> dict[str, Any]:
+    ok, reason = validate_job_request(job)
+    if not ok:
+        return {
+            "status": "blocked",
+            "stage": "policy",
+            "reason": reason,
+            "published": False,
+            "evaluation_report": None,
+        }
+
+    target_metric = float(job.get("target_metric", 0.8))
+    simulated_metric = float(job.get("simulated_metric", target_metric))
+
+    stages = [
+        {"name": "data_prepare", "status": "ok"},
+        {"name": "synthetic_refine", "status": "ok"},
+        {"name": "train_distill", "status": "ok"},
+        {"name": "evaluate", "status": "ok" if simulated_metric >= target_metric else "fail"},
+        {"name": "regression_test", "status": "ok" if simulated_metric >= target_metric else "fail"},
+    ]
+
+    report = {
+        "created_at": _now(),
+        "metric": simulated_metric,
+        "target_metric": target_metric,
+        "passed": simulated_metric >= target_metric,
+        "notes": job.get("evaluation_notes", ""),
+    }
+
+    if not report["passed"]:
+        return {
+            "status": "failed_evaluation",
+            "stage": "evaluate",
+            "stages": stages,
+            "published": False,
+            "evaluation_report": report,
+        }
+
+    return {
+        "status": "ready",
+        "stage": "complete",
+        "stages": stages,
+        "published": False,
+        "evaluation_report": report,
+    }
+
+
+def publish_artifact(result: dict[str, Any]) -> dict[str, Any]:
+    report = result.get("evaluation_report")
+    if not isinstance(report, dict) or not report.get("passed"):
+        return {
+            "status": "blocked",
+            "reason": "evaluation report missing or not passed",
+            "published": False,
+        }
+
+    out = dict(result)
+    out["status"] = "published"
+    out["published"] = True
+    out["published_at"] = _now()
+    return out